Bol. Soc. Mat. Mex.

(2014) 20:237–255
DOI 10.1007/s40590-014-0038-2

ORIGINAL ARTICLE

Topology of the intersections of quadrics II

Vinicio Gómez Gutiérrez · Santiago López de Medrano

Received: 30 June 2013 / Revised: 25 April 2014 / Accepted: 21 July 2014 /

Published online: 5 September 2014
© Sociedad Matemática Mexicana 2014

Abstract We study the topology of the affine real variety given by the intersection
with the unit sphere of the zero set in Rn of a pair of quadratic forms. We give a
complete topological description of this variety in the generic case: when non-empty,
it is always diffeomorphic to either the unit tangent bundle of a sphere, the product of
two or three spheres, or the connected sum of an odd number of manifolds, each of
them a product of two spheres. With this we complete the partial description given in
(López de Medrano, 1989). Starting with the cases described in that article and other
elementary ones, the proofs are based on the study of three geometric operations that
give new of these varieties from simpler ones. For each operation, the topology of the
new variety can be described, under appropriate conditions, from that of the old one.
The global structure of the proof consists in an elaborate partial ordering of all the
cases, in such a way that at each step those conditions are satisfied.

Keywords Topology of real algebraic varieties · Intersections of quadrics ·

Geometric topology · Connected sums · Open books · Branched coverings

Dedicated to Fico González Acuña on his 70th birthday.

V. Gómez Gutiérrez
Facultad de Ciencias, Universidad Nacional Autónoma de México, Mexico City, Mexico
e-mail: vgomez@ciencias.unam.mx

S. López de Medrano (B)

Instituto de Matemáticas, Universidad Nacional Autónoma de México, Mexico City, Mexico
e-mail: santiago@matem.unam.mx
238 V. Gómez Gutiérrez, S. López de Medrano

Mathematics Subject Classification 14P25 · 57R65 · 57S25

1 Introduction

We study the topology of the affine real variety V = F −1 (0) in Rn , where F : Rn →

R2 is given by two quadratic forms. Due to homogeneity, it is enough to study the
intersection of V with the unit sphere, so, we will be dealing with the variety Y given
by equations

ai j xi x j = 0
bi j xi x j = 0
n
 xi2 = 1.
i=1

where ai j , bi j are the entries of real symmetric n × n matrices A, B.

We will give a complete topological description of this variety in the generic case,
i.e., when the above system of equations is of rank 3 in Y and therefore Y is smooth.
We show that a smooth Y , when non-empty, is always diffeomorphic to either:
(a) The unit tangent bundle of a sphere, or
(b) The product of two spheres, or
(c) The product of three spheres, or
(d) The connected sum of an odd number of manifolds, each of them a product of
two spheres.
The case to which a specific Y corresponds, as well as the dimensions of the spheres
and, in case (d), the number of summands, can be read from a normal form to which
the couple of quadratic maps can be deformed.
This study was started by Wall [8] where he gave the normal form in general
and, in the diagonal case (i.e., when the two quadratic forms are simultaneously
diagonalizable), the homology groups of Y and the diffeomorphism type in case (c).
In January 1984, the second-named author of this paper began to look at this prob-
lem, starting from a question by Alberto Verjovsky. Without knowing the work of
Wall, he rediscovered his normal form and results in the diagonal case and went fur-
ther to describe the diffeomorphism type of Y in most of the diagonal cases of type
(d). This result was obtained along that year and on the next one the details of the
proof were completed. The result was announced in various meetings along 1986 and
finally published in Topology of the Intersection of Quadrics [6], of which the present
article is a natural sequel.
The topological description was still lacking for:
(i) the unsettled diagonal cases,
(ii) all the non-diagonal ones and
(iii) the intersections of three or more quadrics.
Though much work around these varieties was done in other directions, nothing new
was obtained about those open questions until in [1] several new roads were opened
Topology of the intersections of quadrics II 239

towards the understanding of (iii). In [3], these roads were developed to obtain several
new results, thanks to the combination of two different approaches to the study of the
same type of objects that had coexisted for a long time without any contact between
them.
Following the impulse and experience of these developments, new results have been
obtained, including two old dreams of ours: the topological description in cases (i)
and (ii) above (which we give in the present article) and the answer to new questions
by Alberto. A survey of results recently obtained can be found in [7]1 while new lines
of research are being explored.
In Sect. 2, we recall briefly the normal form for the equations of Y and state the
Main Theorem of this article. We give a brief proof of its simplest cases, where there
is a special elementary proof, while the proof of the other cases of the theorem will
occupy the rest of the article.
In Sect. 3, we recall basic facts about group actions and polytopes associated to the
diagonal cases.
In Sect. 4, we describe the geometry of three operations producing new varieties
from old. At the end of each description, we give examples which are, in fact, proofs
of different cases of the Main Theorem.
In Sect. 5, we give the homology groups in all cases, recalling it in the diagonal
cases and deducing the general case through computations of the homology of the
branched coverings that appear in the second operation.
In Sect. 6, we show that the first two operations preserve connected sums of sphere
products. Together with the homology computations and the examples proved in the
first three sections this gives the complete proof of the Main Theorem.
Results, ideas, suggestions and comments by Francisco González Acuña, Fico,
form a dense set throughout the geometric proofs in this article and forthcoming ones,
to which it is impossible to give credit at every point. Instead, the whole article is
dedicated to him.
We thank the referees for pointing to us some passages that needed to be clarified
and for suggestions for the continuation of our work.
This work includes part of the Ph.D. thesis of the first-named author. Some of
the results were proved, while the second-named author occupied the Lluís Santaló
temporal position at the Centre de Recerca Matemàtica in Bellaterra, Barcelona.

2 The normal form and the Main Theorem

Assume now that the above system is regular, so that Y is smooth. If we deform
the coefficients of the equation without breaking the regularity condition then the
diffeomorphism type of Y does not change. The description of the differentiable type
of the manifold Y passes through the process of making such a deformation until we
reach the simplest form:
By a small perturbation, one can assume that the matrices A, B are invertible and
that all the eigenvalues of C := A−1 B are different. Then, C can be diagonalized over

1 As well as a list of many fields where these varieties appear as interesting examples.
240 V. Gómez Gutiérrez, S. López de Medrano

the complex numbers, with respect to a basis of eigenvectors that are orthogonal with
respect to both quadratic forms A and B. Over the real numbers, the two forms will split
into common orthogonal subspaces of dimensions 1 and 2 corresponding, respectively,
to the real and complex eigenvalues. The two-dimensional ones, depending essentially
on a pair of conjugate complex numbers, can be deformed always into the unique form
(u 2 − v 2 , 2uv) without changing the regularity of the system or the differentiable type
of Y .
With respect to the new coordinates, the quadratic forms will be
 
r
i=1 ai xi2 +  sj=1 u 2j − v 2j
r
i=1 bi xi2 +  sj=1 2u j v j .

The fact that we have a regular system is expressed now by the condition that 0 ∈ R2
should not be a convex combination of any two of the λi = (ai , bi ), a property known as
weak hyperbolicity2 (which is clearly generic) and the diffeomorphism type of Y is not
changed if we move the points λi around R2 without breaking this condition. With this
type of deformation, one can join together as many of them as possible in single points
with multiplicity, then push them radially until they are in the unit circle and finally
distribute them evenly along the circle. Then, the λi can be assumed to be (2 + 1)th
roots of unity ρ j , where ρ has argument 2π/(2 + 1) and ρ j appears as coefficient
with multiplicity n j > 0, j = 1, . . . , 2 + 1, so that r = n 1 + n 2 + · · · + n 2+1 . The
number of terms in this partition of r is always odd and the type of Y does not change
if one permutes cyclically the partition, so, we must actually think of a cyclic partition,
i.e., an equivalence class under cyclic permutations (for details, see [6]). This puts the
diagonal part of our pair of quadratic forms in a normal form.
So, the normal form is completely described by the non-negative integers r and s
and, if r > 0, the cyclic partition r = n 1 + n 2 + · · · + n 2l+1 . The differentiable type
of Y will depend only on the sums of  consecutive numbers in the partition:

di = n i + · · · + n i+−1 , i = 1, . . . , 2 + 1

where j in n j has to be reduced modulo 2 + 1 if j > 2 + 1. Now, we can state our

Main Theorem Let Y be the variety corresponding to the non-negative integers r, s

and, when r > 0, the cyclic partition r = n 1 + n 2 + · · · + n 2l+1 . Then,
(i) if either r = 0 and s ≤ 1, or r > 0 and  = s = 0, Y is empty,
(ii) if r = 0 and s > 1, Y is diffeomorphic to the unit tangent bundle of S s−1 ,
(iii) if r > 0,  = 0 and s > 0, Y is diffeomorphic to the product S s−1 × Sr +s−2 ,
(iv) if r > 0, = 1, s = 0, Y is diffeomorphic to the product

S n 1 −1 × S n 2 −1 × S n 3 −1

2 This property was first defined in [2]. The concept was implicited in [8] and independently discovered in
[6]. Chaperon’s well-chosen terminology was soon adopted in the following papers on the subject (among
others [1,3]).
Topology of the intersections of quadrics II 241

(v) and if r > 0 and l + s > 1, Y is diffeomorphic to the connected sum

2+1  d +s−1 
# i S × Sr −di +s−2 .
i=1

To illustrate this theorem, observe that the connected surfaces that are so obtained
are exactly those of genus 1 (for r = 1 and s = 2), genus 3 (for r = 3,  = 1 and
s = 1) and genus 5 (for r = 5,  = 2 and s = 0). Some infinite families of genera of
connected surfaces that can be realized by intersecting more homogeneous quadrics
appear in [4].
Case (i) is immediate and in case (ii), the normal form coincides with the well-
known complex equations of the unit tangent bundle of S s−1 . It is easy to see that in
case (iii), one gets the sphere bundle of the stabilized tangent bundle of the sphere,
which is trivial. Case (iv) follows by a direct manipulation of the equations in the
normal form into three equations of spheres with separate variables.3 The proof of
case (v) will occupy the rest of this article.

3 Actions and polytopes

When s = 0, the pair of quadratic forms is simultaneously diagonalizable, so, the

variety Y is given by
r
 λi xi2 = 0 (1)
i=1
r
 xi2 = 1. (2)
i=1

λi ∈ R2 , i = 1, . . . , r

The manifold Y admits a Zr2 action by changing the signs of the coordinates. The
quotient can be identified with the intersection of Y with the first orthant of Rr and
Y can be reconstructed from this intersection by reflecting it iteratively on all the
coordinate hyperplanes.
By a simple change of coordinates ri = xi2 , this quotient–intersection can be
identified with the d-dimensional convex polytope P given by
r
 λi ri = 0.
i=1
r
 ri = 1.
i=1
ri ≥ 0.

The weak hyperbolicity condition is equivalent to the fact that P is a simple polytope,
meaning that each vertex is exactly in d facets of P.

3 Statements (ii) and (iii), as well as a different proof of (iv), appear in [8]. Case (v) for s = 0 with some
restrictions was proved in [6].
242 V. Gómez Gutiérrez, S. López de Medrano

As a first example, consider the case s = 0, 5 = 1+1+1+1+1. By symmetry, P

touches every coordinate hyperplane ri = 0, so, Y is connected and P is a pentagon.
Then, the Euler characteristic of Y can be computed from its cell decomposition given
by P, all its faces and all their reflections on the coordinate hyperplanes. As a result,
Y is the surface of genus 5.

4 Geometric operations

We will be using three geometric operations that associate to one of our varieties a
different one4 :
(a) For r > 0, the operation Y → Y  which increases by 1 one of the n i and therefore
also the dimension of the variety. This operation depends on the choice of i, but
we will not need to specify this in the notation.
(b) The operation Y → Ỹ which increases s by 1, thus increasing the dimension of
the variety by 2.
(c) For s = 0, the operation of truncating the quotient polytope Y/Zn2 and construct-
ing from the result a new variety Y ∧ of the same dimension. This operation adds
one variable and one equation, so applied to Y it would get us out of our field of
study. Instead, we will apply it to a variety Z given by just one quadratic form,
to obtain one of our varieties Y .
We now proceed to describe more precisely each of these operations in terms of
equations and their geometric versions.

4.1 Operation Y → Y 

It adds a new variable x0 to which we assign an old pair of coefficients, which we can
assume to be the first one a1 , b1 , thus increasing n 1 , r and the dimension of the variety
by 1 each.
Let Y be given, in short notation, by

a1 x12 + G 1 (y) = 0
b1 x12 + G 2 (y) = 0
x12 + |y|2 = 1

(where y ∈ Rn−1 includes all variables other than x1 ).

Then, Y  is given by

a1 (x02 + x12 ) + G 1 (y) = 0

b1 (x02 + x12 ) + G 2 (y) = 0
x02 + x12 + |y|2 = 1.

4 These operations can be defined for intersections of any number k homogeneous quadrics and the unit
sphere, but are only used in this article for k = 2 (a and b) and k = 1 (c).
 Topology of the intersections of quadrics II 243

From the form of these equations, there is an action of S 1 on Y  by rotation in the

coordinate plane (x0 , x1 ), whose quotient can be identified with the submanifold with
boundary

Y+ = Y ∩ {x1 ≥ 0}.

Let also:

Y0 = Y ∩ {x1 = 0}
Y+ = xY ∩ {x0 ≥ 0}.

so, we have a retraction : Y  → Y+ given by (x0 , x1 , y) → ( x02 + x12 , y) that restricts
to retractions Y+ → Y+ and Y → Y+ . This gives an open book decomposition of Y 
whose monodromy is trivial because it comes from an action of S 1 on Y  , so, we have:

Proposition 1 (i) Y  is an open book with binding Y0 , page Y+ and trivial mon-
odromy.
(ii) Y+ is a deformation retract of Y+ and Y+ is diffeomorphic to Y+ × D 1 .
(iii) The inclusion of Y in Y+ as its boundary is homotopic to the retraction Y → Y+ .

Example 1 In the case s = 0, r = 5 with partition 1 + 1 + 1 + 1 + 1, we have seen

that Y is the surface of genus 5 (end of Sect. 3) and that Y0 = S 1 × S 0 × S 0 , since its
partition is 4 = 2 + 1 + 1 (see the end of Sect. 2). It follows that Y+ (being a retraction
of Y , having Y0 as boundary and Y as double) can only be a torus minus four disks.
It is easy to see that Y+ = Y+ × D 1 is the connected sum along the boundary of 5
copies of S 1 × D 2 and Y  = #5 (S 1 × S 2 ).
By induction, when s = 0, r ≥ 5 with partition = (r − 4) + 1 + 1 + 1 + 1, we
have Y = #5 (S 1 × Sr −4 ). This will also follow from Sect. 4.3.

4.2 Operation Y → Ỹ

It adds two new variables (u, v) and thus increases s by 1 and the dimension of the
variety by 2:
Let Y be given, in short notation, by equations in Rn

F1 (x) = 0
F2 (x) = 0
|x|2 = 1.

then, Ỹ is given by equations

F1 (x) + u 2 − v 2 = 0
F2 (x) + 2uv = 0
|x|2 + u 2 + v 2 = 1.
244 V. Gómez Gutiérrez, S. López de Medrano

Proposition 2 (i) Y is a double cover of S n−1 , ramified over Y .

(ii) There is an embedding Y  ⊂ Ỹ such that (Y  ) is a manifold with boundary
diffeomorphic to Y+ ⊂ S n−1 .
 is diffeomorphic to the union of to copies of the exterior of Y+ in S n−1 glued
(iii) Y
by a diffeomorphism of their boundaries.
Proof The mapping in (i) is given by

(x, u, v) = x/|x|

We can also write the equations of Y in complex notation

F(x) + w2 = 0
|x|2 + |w|2 = 1

where F(x) = F1 (x) + i F2 (x) and w = u + iv.

) and simple computations
Clearly,  is well defined (since x cannot be zero at Y
show that
– the inverse image of every point in the sphere consists of two points except for the
points in Y where it is just one.
– Outside Y ,  is a local diffeomorphism.
–  is equivalent to the mapping

{(x, w) : F(x) + w 2 = 0, |x|2 = 1} → {(x, z) : F(x) + z = 0, |x|2 = 1}

given by (x, w) → (x, w2 ), the equivalence given by diffeomorphisms φ(x, w) =

|x| (x, w) and ψ(x) = (x, −F(x)), which proves (i).
1

0 = Y
For (ii) let Y  ∩ {v = 0}. We can identify this manifold with Y  if we add the
0 under
variable u to the collection of variables with coefficient (1, 0). The image of Y
 is the manifold with boundary given by
 
x ∈ S n−1 : F1 (x) ≤ 0, F2 (x) = 0

and can be identified with Y+ .

 is the union of two manifolds with boundary
As for (iii), clearly Y

 ∩ {v ≥ 0}
W := Y
 ∩ {v ≤ 0}
Y

whose intersection is their boundary. They are both diffeomorphic to the exterior
of (Y0 ) and are interchanged by the diffeomorphism (x, w) → (x, −w). If this
diffeomorphism restricted to its boundary could be extended to one of W with itself
one would conclude that Y  would be the double of W and the proof of our Main
Theorem would be enormously simplified. Since we have not been able to prove this
in general, we will take a long detour to arrive at the theorem. 

Topology of the intersections of quadrics II 245

:
Example 2 In simple cases, this is enough to determine Y

– Consider the case s = 0, r = 1 + p + q and assume p ≤ q. Then, Y = S 0 ×

S p−1 × S q−1 . One can take Y+ = S 0 × S p−1 × D q which inside the ( p + q)-
sphere is standard. So, one can assume that Y+ is contained in an equator and the
reflection on this equator extends the glueing diffeomorphism for Y  to the exterior
W . Therefore, Y is the double of W and, the latter being S p+q minus two copies
of S p−1 × D q+1 , Y is (S 1 × S p+q−1 )#2(S p × S q ).
– Cases where Y has dimension 2 not covered so far are s = 0, r = 1 + 1 + 1 + 1 + 1
and s = 1, r = 1 + 1 + 1 for which we know Y is the surface of genus 5 and
3, respectively. Y+ will be in both cases a connected surface with non-empty
boundary and therefore Y+ = Y+ × D 1 is a connected sum along the boundary of
copies of S 1 × D 2 . By a theorem of Hosokawa and Kawauchi [5], this means that
the surfaces are unknotted and Y  is the double cover of S 4 over a standard surface.
The symmetry argument in the previous paragraph shows that Y  is the connected
sum of copies of S 2 × S 2 (5 and 3, respectively).

These results cover all the varieties Y of dimension not bigger than 4, with the
exception of the case 7 = 1 + 1 + 1 + 1 + 1 + 1 + 1, which will be addressed by the
third operation.

4.3 Operation Z → Z ∧

Let Z be given by two equations in R p+q+2 :

p q
 xi2 −  yi2 = 0
i=0 j=0
p q
 xi2 +  yi2 = 1.
i=0 j=0

By elementary manipulations of the equations, it is evident that Z is diffeomorphic

p+q+2
to the product S p × S q and that the corresponding polytope P = Z /Z2 is the
p+q+2
product of two simplices
×
given by the corresponding equations in R+
p q :


p 
q
ti − sj = 0
i=0 j=0
p q
ti + sj = 1
i=0 j=0

Now, we consider the operation of cutting P in two pieces U and V by an hyperplane

H which we assume does not contain any of the vertices of P, so that both pieces are
also simple polytopes. We will construct varieties Y (U ), Y (V ) which have them as
246 V. Gómez Gutiérrez, S. López de Medrano

polytopes5 so that Z = Y+ (U ) ∪ Y+ (V ). The general idea6 is that if Y+ (U ) is simple

enough we should be able to describe its complement Y+ (V ) in Z and therefore the
double of Y+ (V ) which is Y (V ). Details in our cases follow:
(A) Cutting faces. For 0 ≤ b < q,  > 0 let H ⊂ R p+q be the hyperplane given
by the equation

p 
q
ti + sj = 
i=1 j=b+1
Let U be the intersection of P with the half-space

p 
q
ti + sj ≤ 
i=1 j=b+1
and V the intersection of P with the half-space

p 
q
ti + s j ≥ .
i=1 j=b+1
If  is sufficiently small, U will be a neighborhood of the face (1/2, 0, . . . , 0) ×
b ⊂

p ×
q . So, U is equivalent to
b ×
p+q−b . This product has p + q + 2 facets:
the new facet contained in H and the others are subsets of the facets of P.
We can represent U and V as the polytopes of intersections of quadrics as follows:
For U , introduce a new variable u and an equation joining it with the previous one:

p 
q
−u = ti + sj − 
i=1 b+1
p+1
to obtain a system that clearly represents U inside R+ × R+ × Rq+1 :

p 
q
u+ ti + sj = 
i=1 j=b+1

p q
ti − sj = 0
i=0 j=0
p q
ti + s j = 1.
i=0 j=0

This system is not of the form of that of a polytope associated to an intersection of

quadrics (see Sect. 3), but we could combine adequately the equations to put the system
in that form and obtain the equations of an intersection of quadrics Y (U ) ⊂ R p+q+3
whose polytope is U and whose normal form can be easily identified (for the details,
see below for the case of V ). From this, we can describe the diffeomorphism type of
Y (U ) and Y+ (U ). But in this case, it is simpler to argue geometrically as follows: Y (U )

5 It can be shown that any simple polytope can be realized as the associated polytope of an intersection of
diagonal quadrics, in general more than two, but here the construction will be explicit.
6 This idea was used in [3] to find the topology of some intersections of more than two quadrics.
Topology of the intersections of quadrics II 247

is obtained from U =
b ×
p+q−b by reflecting it in all the p + q + 3 coordinate
hyperplanes. But U has only p + q + 2 facets so, it does not touch all the coordinate
hyperplanes. If we reflect U in all the hyperplanes that touch it (i.e., in all its facets),
we obtain S b × S p+q−b , but we have still to reflect in the remaining hyperplane that
does not intersect it. Therefore, Y (U ) consists of two copies of S b × S p+q−b . If we
consider only the part Y+ (U ) with x0 ≥ 0 that is if we reflect on all hyperplanes except
u = 0 we clearly obtain two copies of S b × D p+q−b .
Now, we look at V . We introduce again a new variable u and an equation joining it
with the previous one:

p 
q
u= ti + s j − .
i=1 b+1
p+1 q+1
So, we can describe V by the following equations in R+ × R+ × R+ :

p 
q
u− ti − s j = −
i=1 j=b+1

p q
ti − sj = 0
i=0 j=0
p q
ti + s j = 1.
i=0 j=0
If we add to the first equation the third one multiplied by , we get:

p 
b 
q
u + t0 + (−1 + )ti + s j + (−1 + )s j = 0.
i=1 j=0 j=b+1
If to the third one we add a small multiple of the latter we obtain an equation with all the
positive coefficients. We can substitute this one with the equation of the unit simplex
p+1 q+1
in R+ × R+ × R+ without changing the combinatorial type of the polytope. So,
we can write V by the system:


p 
b 
q
u + t0 + (−1 + )ti + s j + (−1 + )s j = 0
i=1 j=0 j=b+1
p 
q
ti − sj = 0
i=0 j=0
p q
u+ ti + s j = 1.
i=0 j=0

The coefficients of this system are the following, where we have marked in Fig. 1
their multiplicities:
1. (1, 0) with multiplicity 1 (corresponds to the variable u).
2. (, 1) with multiplicity 1 (variable t0 ).
248 V. Gómez Gutiérrez, S. López de Medrano

Fig. 1 The configuration

corresponding to Z (V )

3. (−1 + , 1) with multiplicity p (variables t1 , . . . , t p ).

4. (−1 + , −1) with multiplicity q − b (variables sb+1 , . . . , sq ).
5. (, −1) with multiplicity b + 1 (variables s0 , . . . , sb ).

Therefore, we have a configuration of coefficients that can clearly be put in the

standard form 1 + 1 + p + (q − b) + (b + 1).
But, on the other hand, we have that Y (V ) is the double of Y+ (V ) and Y+ (V ) ∪
Y+ (U ) = Z = S p × S q . So, we can describe Y (V ) as follows: consider S p × S q and
remove from it two copies of the interior of S b × D p+q−b . Now, take the double of
this manifold with boundary.
In some cases (and probably in all), we can describe completely this manifold:
recall that b < q and assume now that b < p. Then, the core spheres S b × {0} are
contractible in S p × S q . Furthermore, since 2b + 1 < p + q, they are isotopic to a
pair of unknotted and unlinked spheres inside a ball in S p × S q , so Y+ (V ) can be
thought of as the connected sum of S p × S q with the sphere S p+q minus the tubular
neighborhood of two disjoint standard spheres. Then, it is easy to see that its double,
Y (V ), is diffeomorphic to the connected sum

Y (V ) = 2(S p × S q )#2(S p+q−b−1 × S b+1 )#(S p+q−1 × S 1 ).

Example 3 Consider the case s = 0, r = 1 + 1 + n 3 + n 4 + n 5 where we can assume

n 3 ≥ n 5 (inverting the order of the coordinates in the normal form turns the coefficients
into their conjugates). Then, take p = n 3 , b = n 5 − 1 and q = n 4 + n 5 − 1 and all
the conditions of the above result are satisfied so this case is included.

This proves all non-simply connected cases with  = 2 of the Main Theorem (and
proves again that the surface of genus 5 appears as the case r = 5 with partition
1 + 1 + 1 + 1 + 1).
(B) A deeper cut. Now, we want to build-up the variety Y corresponding to the
primitive case 7 = 1 + 1 + 1 + 1 + 1 + 1 + 1 given by equations in normal form with
Topology of the intersections of quadrics II 249

coefficients (ai , bi ) which are the components of ρ i , ρ 7 = 1.

17 ai xi2 = 0
17 bi xi2 = 0
17 xi2 = 1.

Z will now be the variety given by the two last equations restricted to x7 = 0, and
the corresponding polytope is P =
2 ×
2 :

16 bi ti = 0
16 ti = 1

Let now H be given by

16 ai ti = 0

so, we are in the situation above with p = q = 2 and PH− is the polytope corre-
sponding to Y .

PH− .

While PH+ has coefficients (ai , bi ), i = 1, . . . , 6 and (−1, 0), corresponding to the
case 7 = 3 + 1 + 1 + 1 + 1:
250 V. Gómez Gutiérrez, S. López de Medrano

PH+ .

Therefore,

Y PH+ = 5(S 1 × S 3 )

and Y+ (PH+ ) is the connected sum along the boundary of 5 copies of S 1 × D 3

Since Z (P) = S 2 × S 2 is four-dimensional and simply connected, Y+ (PH+ ) can be
deformed into an open disk D 4 ⊂ S 2 × S 2 and assumed to be standard there. So,

Y+ PH− = (S 2 × S 2 ) \ Y+ PH+ = (S 2 × S 2 )#(S 4 \  5(S 1 × D 3 )).

The second summand is then the connected sum along the boundary of five copies
of S 2 × D 2 and therefore

Y+ PH− = (S 2 × S 2 )# (S 2 × D 2 )
5

Its double is clearly

Y = #7 (S 2 × S 2 )

With this we have proved all four-dimensional cases of our Main Theorem.

5 Homology

Theorem 1 Let Y corresponds to the data 0 < r = n 1 + n 2 + · · · + n 2+1 and s.

Then, the homology of Z coincides with that of the connected sum7

#i=1 (S di +s−1 × S n−di +s−2 ).

2+1

Observe that the number of summands in the connected sum only depends on ,
and so, even if the Betti numbers of Y depend on the numbers n i , their sum does not.

7 See Sect. 2 for the definition of d . Observe that when s = 0,  = 1, Y is not even homotopy equivalent
i
to the connected sum in the theorem, but it does have the same homology groups.
Topology of the intersections of quadrics II 251

We will call the sum of the Betti numbers of a space X its total amount of homology8
and denote it by β(X ).
The diagonal case s = 0 was proved in [6,8] with totally different arguments. So,
we consider this case proved.
For the general case, it is natural to try to start with cases s = 0 and go up inductively
using the geometry of the Y  construction and the Mayer–Vietoris exact sequence of
 + −
the triple (Y ; Y , Y ). But this requires the knowledge of some homology properties
of the glueing diffeomorphism, a fact that we do not know for the moment how to
prove directly. Instead, we will take a detour which at the end of the proof will imply
that fact. An important step will be to prove first that the total amount of homology is
preserved by the operations Y → Y  and Y → Ỹ :

Proposition 3 (Preservation of the total amount of homology) The operations Y →

Y  and Y → Ỹ preserve the total amount of homology.

Proof For the operation Y  , this is clear in the known case s = 0 and is easy to prove
in general:
From the retraction (recall Sect. 4.1) Y → Y+ , it follows that we have an isomor-
phism:

Hi (Y ) ∼
= Hi (Y+ ) ⊕ Hi (Y, Y+ ).

Since

Hi (Y, Y+ ) ∼
= Hi (Y− , Y0 ) ∼
= H n−i (Y− )

(by the excision isomorphism and Lefschetz duality, where n is the dimension of Y ),
we have

Hi (Y ) ∼
= Hi (Y+ ) ⊕ H n−i (Y+ ).

The same applies to Y  :

Hi (Y  ) ∼
= Hi Y+ ⊕ H n+1−i Y+ .

Since we know that Y+ and Y+ are homotopy equivalent, the total amount of
homology of Y and of Y  is the same. 


For the operation Y → Ỹ , it is not easy to prove it directly from a Mayer–Vietoris

argument, except in the case where Y is highly connected:

Lemma 1 Assume Y has dimension 2 p for p ≥ 0, Hi (Y ) = 0 for i = 1, . . . , p − 1

and H p (Y ) is free of even rank 2k. Then, Hi (Ỹ ) = 0 for i = 1, . . . , p and H p+1 (Ỹ )
is free of rank 2k.

8 Sometimes this sum is called the complexity of the space.

252 V. Gómez Gutiérrez, S. López de Medrano

Proof If p = 0 we are in the cases s = 0, 3 = 1 + 1 + 1 or s = 1, r = 1, where

Y = S 0 × S 0 × S 0 (Y = S 0 × S 0 , respectively) and the double cover of S 2 ramified over
Y is well known to be the surface of genus 3 (genus 1, respectively), the Lemma is true
for these cases. For p > 0, it follows from the previous formulas that Hi (Y+ ) = 0 for
i = 1, . . . , p−1 and H p (Y+ ) is free of rank k and that Hi (Y  ) = 0 for i = 1, . . . , p−1
and H p (Y  ) and H p+1 (Y  ) are both free of rank k.
Let W be the exterior of Y+ in S 2 p+2 . By Alexander duality, Hi (W ) = 0 for
i = 0, p + 1 and H p+1 (W ) is free of rank k.
The Mayer–Vietoris sequence for Ỹ = W ∪Y  W is interesting only in the range:

0 → H p+2 (Ỹ ) → H p+1 (Y  ) → H p+1 (W ) ⊕ H p+1 (W )

→ H p+1 (Ỹ ) → H p (Y  ) → 0

) = 0 for i = 1, . . . , p and, by Poincaré duality that H p+2 (Ỹ ) =

It follows that Hi (Y
0 and H p+1 (Ỹ ) are free. It follows now from the exact sequence that it has rank 2k.
Now, we can finish the proof of the proposition: Y arises from a primitive config-
uration (i.e., one with s = 0 and n i = 1 for all i) by applying the operations Y  and
 a number of times each. Since one can assume all the Y
Y  operations are applied first
(because they commute), it follows inductively from the lemma and the preservation
of the total amount of homology by the operation Y  that Y has the same amount of
homology as that primitive configuration. Now, Y and Y  arise from the same primitive
configuration, so, the proposition is proved. 


With this information, we can now dare into the Mayer–Vietoris sequence in gen-
eral:
Assume that Y is connected and non-empty. Due to the retraction Y → Y+ this
implies that Y+ is connected and so are Y+ and Y  .
Let

ri = βi (Y+ ).

We have r0 = 1, rd = rd+1 = 0.
As shown before,

βi (Y ) = ri + rd−i
βi (Y  ) = ri + rd+1−i

and, therefore,
βi (Y  ) = βi (Y ) + rd+1−i − rd−i (3)
(while β0 (W ) = 1 and βd+1 (W ) = 0).
First, we have the Mayer–Vietoris sequence of the triple (S d+2 ; Y+ × D 1 , W ):
 
Hi+1 (S d+2 ) → Hi (Y  ) → Hi Y+ × D 1 ⊕ Hi (W ) → Hi (S d+2 )
Topology of the intersections of quadrics II 253

For 1 ≤ i ≤ d, this implies Hi (Y  ) → Hi (W ) is surjective (even for i = d + 2).

Second, consider the Mayer–Vietoris sequence of (Y ; W, W )

) → Hi (Y  ) → Hi (W ) ⊕ Hi (W ).
Hi+1 (Y  ) → Hi+1 (W ) ⊕ Hi+1 (W ) → Hi+1 (Y

For i = 1, we get H1 (Y) = 0 and, by Poincaré duality, Hd+1 (Y

) = 0.
Let Si be the image of Hi (Y  ) → Hi (W ) ⊕ Hi (W ) and si its rank. Then, since
the homomorphism into the first summand is surjective, we have si ≥ βi (W ). The
sequence splits into shorter ones

) → Hi (Y  ) → Si → 0.
0 → (Hi+1 (W ) ⊕ Hi+1 (W ))/Si+1 → Hi+1 (Y

So, we have

) + βi (Y  ) − si = 0.
(2βi+1 (W ) − si+1 ) − βi+1 (Y

Adding this equations for i = −1 to d + 1, we obtain

 
) +  d+1 βi (Y  ) −  d+1 si = 0
2 0d+2 βi (W ) − 0d+2 si − 0d+2 βi (Y 0 0

 and Y  is equal to that of Y :

or, since the total amount of homology of both Y

0d+2 (βi (W ) − si ) = 0.

Since each summand βi (W ) − si is non-positive, we conclude that

βi (W ) = si .

Plugging this information into the previous formula, we get

) = βi (Y  ) + βi+1 (W ) − βi (W ).
βi+1 (Y

Using formula (1), we obtain

) = βi (Y ) + rd+1−i − rd−i + βi+1 (W ) − βi (W )

βi+1 (Y

Now, if 1 ≤ i ≤ d, βi (W ) = rd+1−i by Alexander duality [while β0 (W ) = 1 and

βd+1 (W ) = 0]. So, for 1 ≤ i ≤ d − 1, we have

) = βi (Y ).
βi+1 (Y

And we already know that H1 (Y ) = 0 and Hd+1 (Y ) = 0.


So, we have proved that the Betti numbers of Y are as expected and, by induction
that the Betti numbers of a connected Y are equal to those of the connected sum in the
theorem. It remains to prove that that the homology groups are all free and to extend
254 V. Gómez Gutiérrez, S. López de Medrano

the result for the (few) non-connected Y . All this can be done following the above
proof, but we will skip this since we will only need the connected case in the proof
of our Main Theorem, which will in turn imply the full version of the theorem in this
section.

6 Preservation of connected sums

We will show in this section that the operations Y  and Y preserve connected sums of
sphere products.
For Y  , this was proved in [3] in the case s = 0 when Y is simply connected and of
dimension at least 5. The proof applies without change in the general case, with the
same hypotheses. Cases not satisfying those conditions have been proved in previous
sections.
, we start with a geometric version of our previous homology result when Y
For Y
is highly connected.
Proposition 4 (I) Assume that Y has dimension 2 p with p ≥ 0 and is ( p − 1)-
connected, with H p (Y ) free of rank 2k. Then:
(a) Y is diffeomorphic to the connected sum of k copies of S p × S p .
(b) Y  is diffeomorphic to the connected sum of k copies of S p × S p+1 .
(c) Y+ is diffeomorphic to the connected sum along the boundary of k copies of
S p × D p+1 .

(d) Y is p-connected with H p+1 (Y ) free of rank 2k.
(II) Assume that Y has dimension 2 p + 1 with p ≥ 0 and is ( p − 1)-connected, with
H p (Y ) free of rank k. Then:
(a) Y is diffeomorphic to the connected sum of k copies of S p × S p+1 .
(b) Y  is diffeomorphic to the connected sum of k1 copies of S p × S p+2 and k2
copies of S p+1 × S p+1 , with k1 + k2 = k.
(c) Y+ is diffeomorphic to the connected sum along the boundary of k1 copies of
S p × D p+2 and k2 copies of S p+1 × D p+1 .

(d) Y is p-connected with H p+1 (Y ) free of rank k.

Proof The cases with p = 0, 1, 2 of I and p = 0, 1 of II have already been proved

at the end of Sect. 4.2, except for one proved at the end of Sect. 4.3. The other cases
follow from Theorem A1 from [3]9 that gives sufficient conditions for a manifold
with boundary to be a connected sum along the boundary of several manifolds of the
form S i × D i+1 . The hypotheses of this theorem are immediately satisfied, except
for the existence of a basis of the homology of Y+ . One can get disjoint embedded
spheres with stably trivial normal bundle (by the Hurewicz theorem and the Whitney
embedding theorem). To see they are trivial, only in part II we need a proof: If p + 1
is even, then, the type of the bundle is determined by the intersection number of 0
section with itself, but in Y × D 1 , any sphere can be separated from itself by moving
it in the D 1 direction. For p + 1 odd, the existence of a non-trivial bundle would give
a summand of Y with torsion in its homology.

9 This theorem formalizes an idea already used in [6].

Topology of the intersections of quadrics II 255

In both parts (a) follows, Y being the boundary of Y+ and (b) also, since Y  is the
double of Y+ , while (d) was proved in the previous section. 


Now, we can prove all cases of our Main Theorem, starting with the cases with
r > 0,  > 1, of dimension at least 5 and simply connected, which follow by
 and starting from:
induction, applying always first s times the operation Y
When  ≥ 3, the primitive case with s = 0.
When  = 2 and for no two contiguous i, we have n i = 1, the case 8 = 1 + 2 +
1 + 2 + 2, s = 0.
When  = 2, s > 0, n 1 = n 2 = 1, the case 6 = 1 + 1 + 1 + 1 + 2, s = 1.
When  = 1, s > 0, the case 6 = 2 + 2 + 2 with s = 1.
The non-simply connected cases with  = 1, 2 are included in the examples in
Sects. 4.2 and 4.3 and, finally, all cases of dimension at most four were covered in
Sects. 3, 4.2 and 4.3.
We have finished the proof of our Main Theorem.

References

1. Bosio, F., Meersseman, L.: Real quadrics in Cn , complex manifolds and convex polytopes. Acta Math.
197(1), 53–127 (2006)
2. Chaperon, M.: Géométrie différentielle et singularités de systèmes dynamiques. Astérisque vol. 138–
139, p 439. Soc. Mathémat. de France (1986)
3. Gitler, S., López de Medrano, S.: Intersections of quadrics, moment-angle manifolds and connected
sums. Geom. Topol. 17, 1497–1534 (2013). http://www.msp.warwick.ac.uk/gt/2013/17-03/p034.xhtml.
doi:10.2140/gt.2013.17.1497
4. Gómez Gutiérrez, V., López de Medrano, S.: Surfaces as complete intersections. In: Riemann and Klein
surfaces, automorphisms, symmetries and moduli spaces. Contemporary Mathematics, vol. 629, pp.
171–180 (2014). doi:10.1090/conm/629/12584
5. Hosokawa, F., Kawauchi, A.: Proposal for unknotted surfaces in four-spaces. Osaka J. Math. 16, 233–248
(1979)
6. López de Medrano, S.: The topology of the intersection of quadrics in Rn , in Algebraic topology (Arcata
Ca, 1986). In: Lecture Notes in Mathematics, vol. 1370, pp. 280–292. Springer, New York (1989)
7. López de Medrano, S.: Singularities of real homogeneous quadratic mappings. Revista de la Real
Academia de Ciencias Exactas, Fisicas y Naturales. Serie A. Matematicas 108(1), 95–112 (2014).
doi:10.1007/s13398-012-0102-6
8. Wall, C.T.C.: Stability, pencils and polytopes. Bull. Lond. Math. Soc. 12, 401–421 (1980)